Customer Segmentation and personality analysis

Introduction

Customer personality analysis is a crucial tool that businesses can use to gain a deeper understanding of their target customers. By analyzing customer data, businesses can identify customer segments and their unique preferences, motivations, and behaviors. This information can then be used to modify products and services to better suit the needs of each customer segment.

For instance, instead of launching a product and marketing it to all customers in the company's database, businesses can use customer personality analysis to identify which customer segment is most likely to buy the product. They can then tailor their marketing strategies to target that specific segment, resulting in higher conversion rates and a better return on investment.

Overview

Customer personality analysis involves gathering data on factors such as demographics, psychographics, and past behavior. This data is then analyzed to identify patterns and insights that can help businesses make informed decisions about product development, marketing, and customer engagement. By understanding their customers at a deeper level, businesses can build stronger relationships, increase customer satisfaction, and ultimately drive growth.

About the Dataset

The analysis is based on a dataset called "Customer Personality Analysis" (https://www.kaggle.com/datasets/imakash3011/customer-personality-analysis?datasetId=1546318&sortBy=voteCount) which was sourced from Kaggle and contributed by Dr. Omar Romero-Hernandez . The dataset contains a variety of attributes that help to analyze the factors to segemnt customers. This dataset has 2,240 rows of observations and 28 columns of variables. Among the variables, there are 5-character variables and 23 numerical variables.

Columns

People

ID: Customer's unique identifier Year_Birth: Customer's birth year Education: Customer's education level Marital_Status: Customer's marital status Income: Customer's yearly household income Kidhome: Number of children in customer's household Teenhome: Number of teenagers in customer's household Dt_Customer: Date of customer's enrollment with the company Recency: Number of days since customer's last purchase Complain: 1 if the customer complained in the last 2 years, 0 otherwise

Products

MntWines: Amount spent on wine in last 2 years MntFruits: Amount spent on fruits in last 2 years MntMeatProducts: Amount spent on meat in last 2 years MntFishProducts: Amount spent on fish in last 2 years MntSweetProducts: Amount spent on sweets in last 2 years MntGoldProds: Amount spent on gold in last 2 years

Promotion

NumDealsPurchases: Number of purchases made with a discount AcceptedCmp1: 1 if customer accepted the offer in the 1st campaign, 0 otherwise AcceptedCmp2: 1 if customer accepted the offer in the 2nd campaign, 0 otherwise AcceptedCmp3: 1 if customer accepted the offer in the 3rd campaign, 0 otherwise AcceptedCmp4: 1 if customer accepted the offer in the 4th campaign, 0 otherwise AcceptedCmp5: 1 if customer accepted the offer in the 5th campaign, 0 otherwise Response: 1 if customer accepted the offer in the last campaign, 0 otherwise

Place

NumWebPurchases: Number of purchases made through the company’s website NumCatalogPurchases: Number of purchases made using a catalogue NumStorePurchases: Number of purchases made directly in stores NumWebVisitsMonth: Number of visits to company’s website in the last month

Problem statement and Objective

Our objective is to create a machine learning model that has the ability to categorize customers. To achieve this, we will be working with a dataset that has been collected from a marketing campaign. The goal of the model is to predict the response of various customer segments to a specific product or service.

Solution and Analysis

Prepare the tools

Import the required tools

import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
import plotly.express as px
import plotly.graph_objects as go
from plotly.subplots import make_subplots
from IPython.display import Markdown
import joblib
import missingno as msno
import numpy as np
from sklearn.preprocessing import StandardScaler
from sklearn.decomposition import PCA
palette = sns.color_palette(['gold','#cc0000', '#ace600','#33cccc'])
from sklearn.pipeline import make_pipeline
from sklearn.compose import ColumnTransformer
from sklearn.preprocessing import OneHotEncoder, OrdinalEncoder, StandardScaler # For handling categorical column and scaling numeric columns
from sklearn.cluster import KMeans
from yellowbrick.cluster import KElbowVisualizer
from sklearn.metrics import silhouette_score
from sklearn.model_selection import train_test_split
from sklearn.metrics import classification_report, confusion_matrix
from sklearn.ensemble import GradientBoostingClassifier
import joblib

Setup plotly

This function is required for plotly to run properly in colab without it sometimes the plots don't render properly.

To avoid code repeatability, enable plotly in cell function is created

def enable_plotly_in_cell():
  import IPython
  from plotly.offline import init_notebook_mode
  display(IPython.core.display.HTML('''<script src="/static/components/requirejs/require.js"></script>'''))
  init_notebook_mode(connected=False)

Load the data

Read dataset from csv file

dataset =  pd.read_csv("marketing_campaign.csv")

Understanding the data

Descriptive Analysis

dataset.shape

(2240, 29)

print(f'We see the dataset has {dataset.shape[0]} observations and {dataset.shape[1]} features.')

We see the dataset has 2240 observations and 29 features.

Review the data and sample data

dataset.head()

	ID	Year_Birth	Education	Marital_Status	Income	Kidhome	Teenhome	Dt_Customer	Recency	MntWines	...	NumWebVisitsMonth	AcceptedCmp3	AcceptedCmp4	AcceptedCmp5	AcceptedCmp1	AcceptedCmp2	Complain	Z_CostContact	Z_Revenue	Response
0	5524	1957	Graduation	Single	58138.0	0	0	04-09-2012	58	635	...	7	0	0	0	0	0	0	3	11	1
1	2174	1954	Graduation	Single	46344.0	1	1	08-03-2014	38	11	...	5	0	0	0	0	0	0	3	11	0
2	4141	1965	Graduation	Together	71613.0	0	0	21-08-2013	26	426	...	4	0	0	0	0	0	0	3	11	0
3	6182	1984	Graduation	Together	26646.0	1	0	10-02-2014	26	11	...	6	0	0	0	0	0	0	3	11	0
4	5324	1981	PhD	Married	58293.0	1	0	19-01-2014	94	173	...	5	0	0	0	0	0	0	3	11	0

5 rows × 29 columns

dataset.sample(10)

	ID	Year_Birth	Education	Marital_Status	Income	Kidhome	Teenhome	Dt_Customer	Recency	MntWines	...	NumWebVisitsMonth	AcceptedCmp3	AcceptedCmp4	AcceptedCmp5	AcceptedCmp1	AcceptedCmp2	Complain	Z_CostContact	Z_Revenue	Response
578	10242	1966	PhD	Together	47472.0	1	1	16-09-2013	39	56	...	5	0	0	0	0	0	0	3	11	0
852	4856	1957	2n Cycle	Married	82347.0	0	0	06-11-2012	38	556	...	3	1	0	0	1	0	0	3	11	0
1252	5153	1967	PhD	Married	77766.0	0	1	22-02-2013	97	1004	...	6	1	0	0	0	0	0	3	11	1
651	6728	1985	Graduation	Together	29103.0	1	0	27-08-2012	38	16	...	9	0	0	0	0	0	0	3	11	0
220	1755	1988	2n Cycle	Together	45204.0	1	0	11-04-2014	93	30	...	3	0	0	0	0	0	0	3	11	0
1146	749	1981	Graduation	Together	42162.0	1	0	19-11-2013	31	46	...	6	1	0	0	0	0	0	3	11	1
829	6646	1984	Graduation	Together	16529.0	1	0	01-03-2014	23	3	...	6	0	0	0	0	0	0	3	11	0
58	8557	1982	Graduation	Single	NaN	1	0	17-06-2013	57	11	...	6	0	0	0	0	0	0	3	11	0
233	9617	1951	Graduation	Single	15033.0	0	0	16-04-2014	37	0	...	3	0	0	0	0	0	0	3	11	0
2065	123	1951	Graduation	Widow	67046.0	0	1	06-01-2013	92	544	...	2	0	0	0	0	0	0	3	11	0

10 rows × 29 columns

Check datatype and information regarding all different features

We want to look at the datatypes and check to see if they were interpreted correctly.

dataset.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 2240 entries, 0 to 2239
Data columns (total 29 columns):
 #   Column               Non-Null Count  Dtype  
---  ------               --------------  -----  
 0   ID                   2240 non-null   int64  
 1   Year_Birth           2240 non-null   int64  
 2   Education            2240 non-null   object 
 3   Marital_Status       2240 non-null   object 
 4   Income               2216 non-null   float64
 5   Kidhome              2240 non-null   int64  
 6   Teenhome             2240 non-null   int64  
 7   Dt_Customer          2240 non-null   object 
 8   Recency              2240 non-null   int64  
 9   MntWines             2240 non-null   int64  
 10  MntFruits            2240 non-null   int64  
 11  MntMeatProducts      2240 non-null   int64  
 12  MntFishProducts      2240 non-null   int64  
 13  MntSweetProducts     2240 non-null   int64  
 14  MntGoldProds         2240 non-null   int64  
 15  NumDealsPurchases    2240 non-null   int64  
 16  NumWebPurchases      2240 non-null   int64  
 17  NumCatalogPurchases  2240 non-null   int64  
 18  NumStorePurchases    2240 non-null   int64  
 19  NumWebVisitsMonth    2240 non-null   int64  
 20  AcceptedCmp3         2240 non-null   int64  
 21  AcceptedCmp4         2240 non-null   int64  
 22  AcceptedCmp5         2240 non-null   int64  
 23  AcceptedCmp1         2240 non-null   int64  
 24  AcceptedCmp2         2240 non-null   int64  
 25  Complain             2240 non-null   int64  
 26  Z_CostContact        2240 non-null   int64  
 27  Z_Revenue            2240 non-null   int64  
 28  Response             2240 non-null   int64  
dtypes: float64(1), int64(25), object(3)
memory usage: 507.6+ KB

msno.matrix(dataset);

Observations

1. There are missing values in income
2. There are some features in the dataset which are of categorical type
3. DT_customer indicates the date when a customer joined however it's not parsed as Date time

Learning the data mathematically

dataset.describe(include = 'all')

	ID	Year_Birth	Education	Marital_Status	Income	Kidhome	Teenhome	Dt_Customer	Recency	MntWines	...	NumWebVisitsMonth	AcceptedCmp3	AcceptedCmp4	AcceptedCmp5	AcceptedCmp1	AcceptedCmp2	Complain	Z_CostContact	Z_Revenue	Response
count	2240.000000	2240.000000	2240	2240	2216.000000	2240.000000	2240.000000	2240	2240.000000	2240.000000	...	2240.000000	2240.000000	2240.000000	2240.000000	2240.000000	2240.000000	2240.000000	2240.0	2240.0	2240.000000
unique	NaN	NaN	5	8	NaN	NaN	NaN	663	NaN	NaN	...	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN
top	NaN	NaN	Graduation	Married	NaN	NaN	NaN	31-08-2012	NaN	NaN	...	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN
freq	NaN	NaN	1127	864	NaN	NaN	NaN	12	NaN	NaN	...	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN
mean	5592.159821	1968.805804	NaN	NaN	52247.251354	0.444196	0.506250	NaN	49.109375	303.935714	...	5.316518	0.072768	0.074554	0.072768	0.064286	0.013393	0.009375	3.0	11.0	0.149107
std	3246.662198	11.984069	NaN	NaN	25173.076661	0.538398	0.544538	NaN	28.962453	336.597393	...	2.426645	0.259813	0.262728	0.259813	0.245316	0.114976	0.096391	0.0	0.0	0.356274
min	0.000000	1893.000000	NaN	NaN	1730.000000	0.000000	0.000000	NaN	0.000000	0.000000	...	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	3.0	11.0	0.000000
25%	2828.250000	1959.000000	NaN	NaN	35303.000000	0.000000	0.000000	NaN	24.000000	23.750000	...	3.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	3.0	11.0	0.000000
50%	5458.500000	1970.000000	NaN	NaN	51381.500000	0.000000	0.000000	NaN	49.000000	173.500000	...	6.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	3.0	11.0	0.000000
75%	8427.750000	1977.000000	NaN	NaN	68522.000000	1.000000	1.000000	NaN	74.000000	504.250000	...	7.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	3.0	11.0	0.000000
max	11191.000000	1996.000000	NaN	NaN	666666.000000	2.000000	2.000000	NaN	99.000000	1493.000000	...	20.000000	1.000000	1.000000	1.000000	1.000000	1.000000	1.000000	3.0	11.0	1.000000

11 rows × 29 columns

Data Pre-processing

Look for missing values

Even though we have identified missing values for income. Let's check to review to if there are any other missing values which need to be handled

dataset.isnull().sum()

ID                      0
Year_Birth              0
Education               0
Marital_Status          0
Income                 24
Kidhome                 0
Teenhome                0
Dt_Customer             0
Recency                 0
MntWines                0
MntFruits               0
MntMeatProducts         0
MntFishProducts         0
MntSweetProducts        0
MntGoldProds            0
NumDealsPurchases       0
NumWebPurchases         0
NumCatalogPurchases     0
NumStorePurchases       0
NumWebVisitsMonth       0
AcceptedCmp3            0
AcceptedCmp4            0
AcceptedCmp5            0
AcceptedCmp1            0
AcceptedCmp2            0
Complain                0
Z_CostContact           0
Z_Revenue               0
Response                0
dtype: int64

We can confirm there are 24 missing values for income feature. There are no missing values for any other feature. We will be handling the missing values for income in an upcoming step

Check for duplicated records

print(f'There are {dataset.duplicated().sum()} duplicated rows in the dataset.')

There are 0 duplicated rows in the dataset.

Rename columns of dataset

This is done to make it easier to process the data. At the same time, it helps with the understanding and analysis of the data

dataset.rename(columns = {'MntGoldProds':'MntGoldProducts'}, inplace = True)

Handling the missing values for Income feature

data = dataset.dropna(axis=0)
data.isnull().sum()

ID                     0
Year_Birth             0
Education              0
Marital_Status         0
Income                 0
Kidhome                0
Teenhome               0
Dt_Customer            0
Recency                0
MntWines               0
MntFruits              0
MntMeatProducts        0
MntFishProducts        0
MntSweetProducts       0
MntGoldProducts        0
NumDealsPurchases      0
NumWebPurchases        0
NumCatalogPurchases    0
NumStorePurchases      0
NumWebVisitsMonth      0
AcceptedCmp3           0
AcceptedCmp4           0
AcceptedCmp5           0
AcceptedCmp1           0
AcceptedCmp2           0
Complain               0
Z_CostContact          0
Z_Revenue              0
Response               0
dtype: int64

Check skewness before imputation

data['Income'].skew()

6.7634873728111184

If the skewness is between -0.5 and 0.5, the data are fairly symmetrical. If the skewness is between -1 and – 0.5 or between 0.5 and 1, the data are moderately skewed. If the skewness is less than -1 or greater than 1, the data are highly skewed.

Replacing null values with median cause the data is skewed which means there are outliers in the data

data['Income'].fillna(data['Income'].median(), inplace = True)

Correcting the datatype

Converting columns to datetime format

data['Dt_Customer'] = pd.to_datetime(dataset['Dt_Customer'])

data['Year_Birth'] = pd.to_datetime(dataset['Year_Birth'], format ='%Y')

Feature engineering

Examine unique values for various categories

data['Education'].value_counts()

Graduation    1116
PhD            481
Master         365
2n Cycle       200
Basic           54
Name: Education, dtype: int64

data['Marital_Status'].value_counts()

Married     857
Together    573
Single      471
Divorced    232
Widow        76
Alone         3
Absurd        2
YOLO          2
Name: Marital_Status, dtype: int64

print("The newest customer's enrolment date in the records:", max(dataset['Dt_Customer']))
print("The oldest customer's enrolment date in the records:", min(dataset['Dt_Customer']))

The newest customer's enrolment date in the records: 31-12-2013
The oldest customer's enrolment date in the records: 01-01-2013

We only have data of three years (2012 to 2014)

Create feature from Dt_Customer that indicates the number of days a customer is registered in the firm's database. To keep it simple, I am taking this value relative to the most recent customer in the record.

data['Date_Collected'] = '01-01-2015'
data['Date_Collected'] = pd.to_datetime(data.Date_Collected)
data['Time_Enrolled_Days'] = (data['Date_Collected'] - data['Dt_Customer']).dt.days

Create feature to extract "Age" of a customer by the "Year_Birth"

data['Age'] = (data["Dt_Customer"].dt.year.max()) - (data['Year_Birth'].dt.year)

Create feature to calculate amount of years the person has been a customer

data['Years_Customer'] = (data["Dt_Customer"].dt.year.max()) - (data['Dt_Customer'].dt.year)

data['Month_Customer'] = 12.0 * (2015 - data["Dt_Customer"].dt.year ) + (1 - data["Dt_Customer"].dt.month)

data['Days_Customer'] = (data["Dt_Customer"].max()) - (data['Dt_Customer'])

Create feature Total money spent to indicate the total amount spent by the customer in various categories.

data['TotalMoneySpent'] = data['MntWines']+data['MntFruits']+data['MntMeatProducts']+data['MntFishProducts']+data['MntSweetProducts']+data['MntGoldProducts']

Create feature for Total number of purchases made to indicate the total number of purchases made by the customer in various categories.

data['TotalNumOfPurchases'] = data['NumWebPurchases'] + data['NumCatalogPurchases'] + data['NumStorePurchases'] + data['NumDealsPurchases']

Create feature for Total number of accepted campagins to indicate the total accpeted campaigns by the customer in various categories.

data['TotalAccCmp'] = data['AcceptedCmp1'] + data['AcceptedCmp2'] + data['AcceptedCmp3'] + data['AcceptedCmp4'] + data['AcceptedCmp5'] + data['Response']

Deriving living situation by marital status

data["Partner"]=data["Marital_Status"].replace({"Married":"Yes", "Together":"Yes", "Absurd":"No", "Widow":"No", "YOLO":"No", "Divorced":"No", "Single":"No","Alone":"No"})

Segmenting education levels in three groups

data["Education"]=data["Education"].replace({"Basic":"Undergraduate","2n Cycle":"Undergraduate", "Graduation":"Graduate", "Master":"Postgraduate", "PhD":"Postgraduate"})

Create feature for Total children living in the house

data["Children"] = data["Kidhome"] +  data["Teenhome"]

To get further clarity of household, Creating feature indicating Family Size

data['Family_Size'] = data['Partner'].replace({'No': 1, 'Yes':2}) + data['Children']

Create a feature Is Parent to indicate parenthood status

data['Is_Parent'] = np.where(data.Children > 0, 1, 0)

Adding columns about the day, month and year cutomer joined

data['YearJoined'] = data['Dt_Customer'].dt.year
data['MonthJoined'] = data['Dt_Customer'].dt.strftime("%B")
data['DayJoined'] = data['Dt_Customer'].dt.day_name()

Dividing age into groups

data['AgeGroup'] = pd.cut(x = data['Age'], bins = [1, 17, 24, 44, 64, 150],
                         labels = ['Under 18','Young adult','Adult','Middel Aged','Senior Citizen'])

Dropping columns which are not required

data.drop(['Z_CostContact','Z_Revenue','Year_Birth','Dt_Customer', 'Marital_Status'], axis=1, inplace=True)

data.head()

	ID	Education	Income	Kidhome	Teenhome	Recency	MntWines	MntFruits	MntMeatProducts	MntFishProducts	...	TotalNumOfPurchases	TotalAccCmp	Partner	Children	Family_Size	Is_Parent	YearJoined	MonthJoined	DayJoined	AgeGroup
0	5524	Graduate	58138.0	0	0	58	635	88	546	172	...	25	1	No	0	1	0	2012	April	Monday	Middel Aged
1	2174	Graduate	46344.0	1	1	38	11	1	6	2	...	6	0	No	2	3	1	2014	August	Sunday	Middel Aged
2	4141	Graduate	71613.0	0	0	26	426	49	127	111	...	21	0	Yes	0	2	0	2013	August	Wednesday	Middel Aged
3	6182	Graduate	26646.0	1	0	26	11	4	20	10	...	8	0	Yes	1	3	1	2014	October	Thursday	Adult
4	5324	Postgraduate	58293.0	1	0	94	173	43	118	46	...	19	0	Yes	1	3	1	2014	January	Sunday	Adult

5 rows × 41 columns

data.columns

Index(['ID', 'Education', 'Income', 'Kidhome', 'Teenhome', 'Recency',
       'MntWines', 'MntFruits', 'MntMeatProducts', 'MntFishProducts',
       'MntSweetProducts', 'MntGoldProducts', 'NumDealsPurchases',
       'NumWebPurchases', 'NumCatalogPurchases', 'NumStorePurchases',
       'NumWebVisitsMonth', 'AcceptedCmp3', 'AcceptedCmp4', 'AcceptedCmp5',
       'AcceptedCmp1', 'AcceptedCmp2', 'Complain', 'Response',
       'Date_Collected', 'Time_Enrolled_Days', 'Age', 'Years_Customer',
       'Month_Customer', 'Days_Customer', 'TotalMoneySpent',
       'TotalNumOfPurchases', 'TotalAccCmp', 'Partner', 'Children',
       'Family_Size', 'Is_Parent', 'YearJoined', 'MonthJoined', 'DayJoined',
       'AgeGroup'],
      dtype='object')

data.shape

(2216, 41)

data['Days_Customer'] = data['Days_Customer'].dt.days.astype('int16')

data['Days_Customer']

0       971
1       125
2       472
3        65
4       321
       ... 
2235    541
2236     61
2237    315
2238    316
2239    782
Name: Days_Customer, Length: 2216, dtype: int16

Exploratory Data Analysis

Examining Unique Values

unique_number = []
for i in data.columns:
    x = data[i].value_counts().count()
    unique_number.append(x)

pd.DataFrame(unique_number, index = data.columns, columns = ["Total Unique Values"])

	Total Unique Values
ID	2216
Education	3
Income	1974
Kidhome	3
Teenhome	3
Recency	100
MntWines	776
MntFruits	158
MntMeatProducts	554
MntFishProducts	182
MntSweetProducts	176
MntGoldProducts	212
NumDealsPurchases	15
NumWebPurchases	15
NumCatalogPurchases	14
NumStorePurchases	14
NumWebVisitsMonth	16
AcceptedCmp3	2
AcceptedCmp4	2
AcceptedCmp5	2
AcceptedCmp1	2
AcceptedCmp2	2
Complain	2
Response	2
Date_Collected	1
Time_Enrolled_Days	662
Age	59
Years_Customer	3
Month_Customer	36
Days_Customer	662
TotalMoneySpent	1047
TotalNumOfPurchases	39
TotalAccCmp	6
Partner	2
Children	4
Family_Size	5
Is_Parent	2
YearJoined	3
MonthJoined	12
DayJoined	7
AgeGroup	5

data.describe(include=object).T

	count	unique	top	freq
Education	2216	3	Graduate	1116
Partner	2216	2	Yes	1430
MonthJoined	2216	12	August	210
DayJoined	2216	7	Sunday	342

Martial Status

enable_plotly_in_cell()

maritalstatus = data['Partner'].value_counts()

fig = px.pie(maritalstatus,
             values = maritalstatus.values,
             names = maritalstatus.index,
             color_discrete_sequence=px.colors.sequential.RdBu)
fig.update_traces(textposition='inside', textinfo='percent+label',
                  marker = dict(line = dict(color = 'white', width = 4)))
fig.show()

Observation

1. Majority of the customers have partners.
2. Only 1/3rd are singles

Spending by Marital Status

enable_plotly_in_cell()
maritalspending = data.groupby('Partner')['TotalMoneySpent'].mean().sort_values(ascending=False)
maritalspending_df = pd.DataFrame(list(maritalspending.items()), columns=['Marital Status', 'Average Spending'])

plt.figure(figsize=(13,5))
sns.barplot(data = maritalspending_df, x="Average Spending", y="Marital Status", palette='rocket');

plt.xticks( fontsize=12)
plt.yticks( fontsize=12)
plt.xlabel('Average Spending', fontsize=13, labelpad=13)
plt.ylabel('Marital Status', fontsize=13, labelpad=13);

Observation

1. Even though the number of customers who have a partner is considerably higher. THe singles spent more money on the average compared to them

Child Analysis

enable_plotly_in_cell()
children = data['Children'].value_counts()

fig = px.pie(children,
             values = children.values,
             names = children.index,
             color_discrete_sequence=px.colors.sequential.RdBu)
fig.update_traces(textposition='inside', textinfo='percent+label',
                  marker = dict(line = dict(color = 'white', width = 2)))
fig.show()

Observation

1. 50% of the customers have 1 child
2. 28% don't have a child and 19% have 2 children

Average Child Wise Spendings

enable_plotly_in_cell()
childrenspending = data.groupby('Children')['TotalMoneySpent'].mean().sort_values(ascending=False)
childrenspending_df = pd.DataFrame(list(childrenspending.items()), columns=['Num of Children', 'Average Spending'])

plt.figure(figsize=(10,5))

sns.barplot(data=childrenspending_df,  x="Num of Children", y="Average Spending", palette='rocket_r');
plt.xticks( fontsize=12)
plt.yticks( fontsize=12)
plt.xlabel('Num of Children', fontsize=13, labelpad=13)
plt.ylabel('Average Spending', fontsize=13, labelpad=13);

sns.pairplot(data , vars=['TotalMoneySpent','Income','Age'] , hue='Children', palette='husl');

Observation

1. Customers who don't have any children at home spent higher than the customers having 1 children. The customers having 1 children are spending higher than the customers havin 2 and 3 children.

Income distribution

plt.figure(figsize=(13,8))
sns.scatterplot(x=data[data['Income']<600000]['TotalMoneySpent'], y=data[data['Income']<600000]['Income'], color='#cc0000');

Observation

1. The salaries of the customers have normal distribution with most of the customers earning between 25000 and 85000.

Age Distribution

enable_plotly_in_cell()
plt.figure(figsize=(10,5))
ax = sns.histplot(data = data['Age'], color='salmon')
ax.set(title = "Age Distribution of Customers");
plt.xticks( fontsize=12)
plt.yticks( fontsize=12)
plt.xlabel('Age ', fontsize=13, labelpad=13)
plt.ylabel('Counts', fontsize=13, labelpad=13);

Observation

1. Age of the customers is nearly normally distributed, with most of the customers aged between 40 and 60.

Age analysis with spending

plt.figure(figsize=(13,8))
sns.scatterplot(x=data['TotalMoneySpent'], y=data['Age']);

Observation

1.There doesn't seem to be any clear relationship between age of customers and their spending habits.

Age group analysis

enable_plotly_in_cell()
agegroup = data['AgeGroup'].value_counts()

fig = px.pie(labels = agegroup.index, values = agegroup.values, names = agegroup.index, width = 550, height = 550)

fig.update_traces(textposition = 'inside',
                  textinfo = 'percent + label',
                  hole = 0.4,
                  marker = dict(colors = ["red", "green", "blue", "goldenrod", "magenta"],
                                line = dict(color = 'white', width = 2)))

fig.update_layout(annotations = [dict(text = 'Age Groups',
                                      x = 0.5, y = 0.5, font_size = 20, showarrow = False,
                                      font_color = 'black')],
                  showlegend = False)

fig.show()

Observation

1. Almost 50% of the customers are adults
2. The second most popular category is middel aged

Age group analysis with spending

agegroupspending = data.groupby('AgeGroup')['TotalMoneySpent'].mean().sort_values(ascending=False)
agegroupspending_df = pd.DataFrame(list(agegroup.items()), columns=['Age Group', 'Average Spending'])

plt.figure(figsize=(20,10))

sns.barplot(data = agegroupspending_df, x="Average Spending", y='Age Group', palette='rocket_r');
plt.xticks( fontsize=16)
plt.yticks( fontsize=16)
plt.xlabel('Average Spending', fontsize=20, labelpad=20)
plt.ylabel('Age Group', fontsize=20, labelpad=20);

Observations

1. Adults spend more than other age groups

Education analysis

data['Education'].value_counts().plot.pie(explode=[0.1,0,0], autopct='%1.1f%%', shadow=True, figsize=(8,8), colors=sns.color_palette("bright"));

enable_plotly_in_cell()

fig = px.histogram(data, x = 'Education',
                   opacity = 1, barmode = 'overlay',
                   width = 800, height = 500,
                  color_discrete_sequence=px.colors.qualitative.G10)

fig.update_layout(title='Education Level')
fig.show()

plt.figure(figsize=(13,8))
sns.histplot(x=data['TotalMoneySpent'], hue=data['Education']);

Observation

1. 50% of caustomers are graduates and only 11% of customers are undergraduates, remaining are postgraduates
2. Majority are well educated

Most Bought Products

products = data[['MntWines', 'MntFruits', 'MntMeatProducts', 'MntFishProducts', 'MntSweetProducts', 'MntGoldProducts']]
product_means = products.mean(axis=0).sort_values(ascending=False)
product_means_df = pd.DataFrame(list(product_means.items()), columns=['Product', 'Average Spending'])

plt.figure(figsize=(15,10))
plt.title('Average Spending on Products')
sns.barplot(data=product_means_df, x='Product', y='Average Spending', palette='rocket_r');
plt.xlabel('Product', fontsize=20, labelpad=20)
plt.ylabel('Average Spending', fontsize=20, labelpad=20);

Observation

1. Wine and Meats products are the most famous products among the customers.
2. Sweets and Fruits are not being purchased often.

Outlier Detection

The presence of outliers in a classification or regression dataset can result in a poor fit and lower predictive modeling performance, therefore we should see there are ouliers in the data

plt.figure(figsize=(13,8))
sns.distplot(data.Age, color='purple');

plt.figure(figsize=(13,8))
sns.distplot(data.Income, color='Yellow');

plt.figure(figsize=(13,8))
sns.distplot(data.TotalMoneySpent, color='#ff9966');

enable_plotly_in_cell()

fig = make_subplots(rows=1, cols=3)

fig.add_trace(go.Box(y=data['Age'], notched=True, name='Age', marker_color = '#6699ff',
                     boxmean=True, boxpoints='suspectedoutliers'), 1, 2)

fig.add_trace(go.Box(y=data['Income'], notched=True, name='Income', marker_color = '#ff0066',
                     boxmean=True, boxpoints='suspectedoutliers'), 1, 1)

fig.add_trace(go.Box(y=data['TotalMoneySpent'], notched=True, name='Money Spent', marker_color = 'lightseagreen',
                     boxmean=True, boxpoints='suspectedoutliers'), 1, 3)

fig.update_layout(title_text='Box Plots for Numerical Variables')

fig.show()

data.head(1)

	ID	Education	Income	Kidhome	Teenhome	Recency	MntWines	MntFruits	MntMeatProducts	MntFishProducts	...	TotalNumOfPurchases	TotalAccCmp	Partner	Children	Family_Size	Is_Parent	YearJoined	MonthJoined	DayJoined	AgeGroup
0	5524	Graduate	58138.0	0	0	58	635	88	546	172	...	25	1	No	0	1	0	2012	April	Monday	Middel Aged

1 rows × 41 columns

numerical = ['Income', 'Recency', 'Age', 'TotalMoneySpent']

def detect_outliers(d):
  for i in d:
    Q3, Q1 = np.percentile(data[i], [75 ,25])
    IQR = Q3 - Q1

    ul = Q3+1.5*IQR
    ll = Q1-1.5*IQR

    outliers = data[i][(data[i] > ul) | (data[i] < ll)]
    print(f'*** {i} outlier points***', '\n', outliers, '\n')

detect_outliers(numerical)

*** Income outlier points*** 
 164     157243.0
617     162397.0
655     153924.0
687     160803.0
1300    157733.0
1653    157146.0
2132    156924.0
2233    666666.0
Name: Income, dtype: float64 

*** Recency outlier points*** 
 Series([], Name: Recency, dtype: int64) 

*** Age outlier points*** 
 192    114
239    121
339    115
Name: Age, dtype: int64 

*** TotalMoneySpent outlier points*** 
 1179    2525
1492    2524
1572    2525
Name: TotalMoneySpent, dtype: int64 

Observation

There are some customers aged above 100. This is unlikely to happen. Let's drop those customers from data.

There are some customers who are earning more than 120,000 and some of them even more than 600,000. They are clearly the outliers in the data, so we will leave them out.

Delete the outlier points

data = data[(data['Age']<100)]
data = data[(data['Income']<600000)]

data.shape

(2212, 41)

Check for Rare Categories

Some categories may appear a lot in the dataset, whereas some other categories appear only in a few number of observations.

Rare values in categorical variables tend to cause over-fitting, particularly in tree based methods. Rare labels may be present in training set, but not in test set, therefore causing over-fitting to the train set. Rare labels may appear in the test set, and not in the train set. Thus, the machine learning model will not know how to evaluate it.

categorical = [var for var in data.columns if data[var].dtype=='O']

# check the number of different labels
for var in categorical:
    print(data[var].value_counts() / np.float(len(data)))
    print()
    print()

Graduate         0.504069
Postgraduate     0.382007
Undergraduate    0.113924
Name: Education, dtype: float64


Yes    0.64557
No     0.35443
Name: Partner, dtype: float64


August       0.094937
October      0.093580
December     0.090416
March        0.089964
January      0.085443
May          0.084539
April        0.083183
February     0.083183
November     0.081374
June         0.075949
September    0.073689
July         0.063743
Name: MonthJoined, dtype: float64


Sunday       0.154611
Wednesday    0.148734
Thursday     0.141953
Tuesday      0.141953
Monday       0.139693
Friday       0.137884
Saturday     0.135172
Name: DayJoined, dtype: float64

As shown above, there is no rare category in the categorical variables.

plt.figure(figsize=(30, 30))
sns.heatmap(data.corr() , annot=True, cmap='coolwarm', linewidths=0.5)
plt.title('Correlation Heatmap')
plt.show()

Observation

1. There are no features which are highly correlated

Feature Transformation

Choosing the features we we will be reporting by removing unnecessary features

subset = data[['Income','Kidhome','Teenhome','Age','Partner','Education']]
print('This is the data we will use for clustering:')
subset.head()

This is the data we will use for clustering:

	Income	Kidhome	Teenhome	Age	Partner	Education
0	58138.0	0	0	57	No	Graduate
1	46344.0	1	1	60	No	Graduate
2	71613.0	0	0	49	Yes	Graduate
3	26646.0	1	0	30	Yes	Graduate
4	58293.0	1	0	33	Yes	Postgraduate

subset.describe()

	Income	Kidhome	Teenhome	Age
count	2212.000000	2212.000000	2212.000000	2212.000000
mean	51958.810579	0.441682	0.505877	45.086347
std	21527.278844	0.536955	0.544253	11.701599
min	1730.000000	0.000000	0.000000	18.000000
25%	35233.500000	0.000000	0.000000	37.000000
50%	51371.000000	0.000000	0.000000	44.000000
75%	68487.000000	1.000000	1.000000	55.000000
max	162397.000000	2.000000	2.000000	74.000000

We aren not scaling the kidhome, teenhome cols, cause their min, max lies between 0 & 2

num_cols = ['Income','Age']
numeric_pipeline = make_pipeline(StandardScaler())

ord_cols = ['Education']
ordinal_pipeline = make_pipeline(OrdinalEncoder(categories=[['Undergraduate','Graduate','Postgraduate']]))

nom_cols = ['Partner']
nominal_pipeline = make_pipeline(OneHotEncoder())

Transform pipelines

transformer = ColumnTransformer(transformers=[('num',numeric_pipeline,num_cols),
                                           ('ordinal', ordinal_pipeline,ord_cols),
                                              ('nominal' ,nominal_pipeline,nom_cols)
                                             ])

transformer

ColumnTransformer(transformers=[('num',
                                 Pipeline(steps=[('standardscaler',
                                                  StandardScaler())]),
                                 ['Income', 'Age']),
                                ('ordinal',
                                 Pipeline(steps=[('ordinalencoder',
                                                  OrdinalEncoder(categories=[['Undergraduate',
                                                                              'Graduate',
                                                                              'Postgraduate']]))]),
                                 ['Education']),
                                ('nominal',
                                 Pipeline(steps=[('onehotencoder',
                                                  OneHotEncoder())]),
                                 ['Partner'])])

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

Fit the transformed data

transformed = transformer.fit_transform(subset)
print('Data has been Transformed')

Data has been Transformed

# subset = data[['Income','Age','Month_Customer','TotalMoneySpent','Children']]
# print('This is the data we will use for clustering:')
# subset.head()

Machine Learning Model

We wii use the Elbow method to identify the optimum clusters

from sklearn.cluster import KMeans

options = range(2,9)
inertias = []

for n_clusters in options:
    model = KMeans(n_clusters, random_state=42).fit(transformed)
    inertias.append(model.inertia_)

plt.figure(figsize=(20,10))
plt.title("No. of clusters vs. Inertia")
plt.plot(options, inertias, '-o', color = 'black')
plt.xticks( fontsize=16)
plt.yticks( fontsize=16)
plt.xlabel('No. of Clusters (K)', fontsize=20, labelpad=20)
plt.ylabel('Inertia', fontsize=20, labelpad=20);

Observation

1. Based on the plot above we will group customers into 4 clusters, because the inertia value does not decrease much after 4 clusters.

Using k-means to form clusters

kmeans = KMeans(n_clusters=4, init='k-means++', random_state=42).fit(transformed)
subset['Clusters'] = kmeans.fit_predict(transformed) #fit the data and adding back clusters to the data in clusters column

subset.head()

	Income	Kidhome	Teenhome	Age	Partner	Education	Clusters
0	58138.0	0	0	57	No	Graduate	3
1	46344.0	1	1	60	No	Graduate	3
2	71613.0	0	0	49	Yes	Graduate	0
3	26646.0	1	0	30	Yes	Graduate	2
4	58293.0	1	0	33	Yes	Postgraduate	1

Count plot to see number of customers in each cluster

plt.figure(figsize=(12, 8))
sns.countplot(x='Clusters', data=subset, palette=palette)
plt.title('Customer Distribution Within Clusters')
plt.show()

Observations

1. Cluster 2 has highest number of customers
2. Cluster 3 has least number of customers

plt.figure(figsize=(12, 8))
sns.countplot(x='Clusters', data=subset, hue='Kidhome', palette=palette)
plt.title('Customer Distribution Children (Kids)')
plt.show()

Clusters Identification

plt.figure(figsize=(12, 8))
sns.countplot(x='Clusters', data=subset, hue='Teenhome', palette=palette)
plt.title('Customer Distribution Children (Teen)')
plt.show()

plt.figure(figsize=(12, 8))
sns.countplot(x='Clusters', data=subset, hue='Partner', palette=palette)
plt.title('Customer Distribution Partner')
plt.show()

plt.figure(figsize=(12, 8))
sns.countplot(x='Clusters', data=subset, hue='Education', palette=palette)
plt.title('Customer Distribution Education')
plt.show()

Observation

1. Kidhome

• Cluster 0 mostly has customers with 1 kid in household
• Cluster 1 has customers with no kids in household
• Cluster 2 also has large number of customers with no kids in household
• Cluster 3 has customers with 0 and 1 kids in household

1. Teenhome:

• Cluster 0 consist of customers with no teen in household & few of them have 1 Teen in household
• Same goes for the cluster 1 & 3
• Cluster 2 has customers with 1 Teen in household

1. Partner

• All the customers in cluster 0 have partner
• All The customers in cluster 3 have no partner
• Cluster 1 & 2 has customers with and without partner, but most of them have partner

1. Education:

• All clusters have customers with graduate, postgraduate and undergraduate background
• All clusters have less number of customers with undergraduate background
• Cluster 2 has highest number of postgraduates and graduates

Find out the customers which have kids in different clusters

plt.figure(figsize=(12, 8))
sns.barplot(x=subset["Clusters"], y=subset["Kidhome"],palette=palette)
plt.title("Kids In Household vs Clusters", size=15)
plt.show()

Observation

1. Cluster 0 and 3 has the maximum number of customers with kids in household
2. Cluster 1 and 2 has the least number of customers with kids in household

Find out the customers which have kids in different clusters

plt.figure(figsize=(12, 8))
sns.barplot(x=subset["Clusters"], y=subset["Teenhome"],palette=palette)
plt.title("Teens In Household vs Clusters", size=15)
plt.show()

Observation

1. Cluster 2 has maximum number of customers having Teens in household
2. Remaining Clusters also have customers with Teens in household but they are less as compared to cluster 2

Findout income of customers with in clusters

plt.figure(figsize=(12, 8))
sns.barplot(x=subset["Clusters"], y=subset["Income"],palette=palette)
plt.title("Income vs Clusters", size=15)
plt.show()

Observations

1. Cluster 1 has high Income followed by cluster 2 even though cluster 2 has highest number of customers and most number of post graduates & graduates as compared to cluster 1
2. Cluster 0 and 3 have least income

Contributing factor to income

plt.figure(figsize=(12, 8))
sns.barplot(x='Clusters', y='Income', data=subset, hue='Education', palette=palette)
plt.title('Cluster Income Distribution with Education')
plt.show()

plt.figure(figsize=(12, 8))
sns.barplot(x='Clusters', y='Income', data=subset, hue='Kidhome', palette=palette)
plt.title('Cluster Income Distribution with Kids')
plt.show()

plt.figure(figsize=(12, 8))
sns.barplot(x='Clusters', y='Income', data=subset, hue='Teenhome', palette=palette)
plt.title('Cluster Income Distribution with Teenhome')
plt.show()

plt.figure(figsize=(12, 8))
sns.barplot(x='Clusters', y='Income', data=subset, hue='Partner', palette=palette)
plt.title('Cluster Income Distribution with Partner')
plt.show()

Observations

1. Interesting thing to note is that the number of customers with 2 kids or teens is very very less still their income is similar to the customers which have no kids or teens or have 1 kid & teen. So I conclude that the customers with more than 1 kid or teen in houehold has high household income 1
2. From the plots for Education and partner, I can conclude it has nothing to do with income. Infact undergraduates are earniing equal or more than graduates and postgraduates within each cluster.

Customer Profiling

1. Cluster 1

• Fewer customers but with the highest income
• No kids, few have 1 teen
• Graduates and Post Graduates
• Most of them have partners

1. Cluster 2

• Max number of customers and high income
• No kids, few have 1 or 2 teens
• High number of post graduates and graduates
• Most of them have partners

1. Cluster 0

They have the least income

• 1 kid and few have teen
• Graduates and post graudates but also has most graduates
• All have partners

1. Cluster 3 Few customers and less income

• 1 kid and few have teen
• Graduates and Post graduates
• All have no partner

Model Building

Separate features and target column

x = subset.drop('Clusters', axis=1)
y = subset['Clusters']

Create train and test data

x_train, x_test, y_train, y_test = train_test_split(x,y, test_size=0.3, random_state=42)

Adding GradientBoostingClassifier to transformer pipeline

final_pipeline = make_pipeline(transformer, GradientBoostingClassifier())

Fit the data to new pipeline & model

final_pipeline.fit(x_train, y_train)

Check the accuracy of our model

final_pipeline.score(x_test, y_test)

joblib.dump(final_pipeline, 'customer_segmentation_cluster.pkl')